My first guess is that he looked at the mesh’s hard-coded polygons through Roblox’s mesh formatting, like if you opened a mesh file with something a bit more complex than Notepad, and created parts based off of the normal and position generated from each triangle. There was an article about how to read Roblox’s mesh format here:
My other guess is that he casted a lot of rays onto a mesh, and using their normals and positions returned, created a part at the ray’s point of contact. If this were the case though, he would have probably not been able to resize parts so that the copied mesh actually looked somewhat correct. There would be a lot of identical parts that would be under/oversized.